mcp-repo-graph

Scan the codebase and (re)build the structural graph. Auto-detects languages and frameworks. Call on first use, after major refactors, or when graph data feels stale.

Repo overview: git state, feature coverage, available flows. Cheap orientation.

End-to-end flow for a feature: entry point through service layer to data, rendered as layered tiers (ENTRY -> SERVICE -> HANDLER -> DATA).

Shortest path between two nodes. Accepts exact IDs or fuzzy name matches. Shows tier transitions along the path.

Blast radius: fan out from a node to see what it affects (downstream) or depends on (upstream). Groups results by architectural tier.

All direct connections to and from a node — one hop in each direction, with edge types.

Total context cost (lines) for a feature's flow. Shows per-file line counts. Use before starting work to know if the feature fits in context.

Files ranked by size x connection count — the biggest maintenance risks. High-coupling large files are the worst context hogs.

Smallest file set needed for a task. Filters by sub_task keywords if given, otherwise returns the full feature file set ranked by relevance.

Internal structure of a file: functions/methods ranked by size, class counts, injected services. Use to understand what's inside before splitting.

Concrete split suggestions for an oversized file, grouped by responsibility and cohesion. Works with any supported language.

Visual ASCII graph map. With feature: layered flow diagram. With node: tree of children, connections, and flows. Without args: full graph overview with node/edge type counts.

Reload graph data from disk after a regeneration.